Enhancing Performance of Biomedical Named Entity Recognition

Abstract

Identifying biomedical entities such as genes, proteins, cell types, and cell lines in the field of named entity recognition has been acknowledged as a challenging task. In this paper, the performance of biomedical named entity recognition has been enhanced by using a filter-based feature selection technique. Various feature types have been considered in the selection process such as morphological, orthographical, part of speech, context, and word embedding. The Chi-squared and ReliefF feature selection techniques have been utilized to reduce the feature vector, which is then taken as an input to structured support vector machine to extract biomedical entities. An innovative approach has been proposed and evaluated on the popular dataset “Genia” using common evaluation metrics. Results revealed that the whole performance of biomedical named entity recognition increased when using the adopted feature selection techniques